model link in experiment setup

Currently, model is set during experiment setup. But this hinder the user to customose the model without changing the code inside MLT.

Solution: Keep model setup as it is, and only move import of model itself to the experiment setup. Load the vanilla model as default if no model is given. All other models need to be loaded in the run script and parsed via a corresponding kwargs to the experiment setup. Do not load the model or compile it, just save the parameter (=pointer) in datastore and load it in the model setup stage.

REMARK: Consider import behaviour, because window_history_size and other parameters are required on initialisation. Is it possible to store the import statement in data store or otherwise create a new object without init and perform init during model setup phase.